Académique Documents
Professionnel Documents
Culture Documents
John A Anderson*
Queensland University of Technology
Abstract: Point and Figure charting is one of the oldest practitioner techniques for
analysing price movements in financial markets, yet has received almost no coverage in
the academic finance literature. This paper empirically contributes to the existing trading
rule literature by providing a methodology for the calculation of Point and Figure charts
using ultra-high-frequency data and tests trading rules using eight objective, pre-defined
trading rules on US futures contracts. In general it is found that profits were able to be
generated during the test period using the Point and Figure methodology.
I INTRODUCTION
Point and Figure charting is a technical analysis technique in which time is not
represented on the x-axis, but merely price changes (independent of time) are recorded
via a series of X’s for increasing price movements and O’s for decreasing price
movements. Evidence suggests that the technique is over 100 years old and is now a
standard feature on many widely-used professional market analysis software systems
such as Bloomberg, Reuters, TradeStation and MetaStock.
Taylor and Allen (1992), surveyed foreign exchange dealers in London about their
analytical techniques and found that over 90% of survey respondents relied on technical
analysis at some point for asset allocation decisions. Given Point and Figure’s place as a
standard feature on popular market analysis software, presumably some of those buy/sell
decisions were made on the basis of Point and Figure techniques although this has not
been specifically documented. Therefore, although we may assume that Point and Figure
does play some role among financial markets practitioners, the academic literature has
left the question of the usefulness of this technique largely ignored.
Point and Figure dispenses with time on the x-axis and concentrates solely on changes in
asset prices, regardless of the time required to produce such price movements. This
means that data, particularly ultra-high-frequency data, can be considerably condensed by
discarding small price changes, while still capturing user-determined levels of significant
price changes on a continuous basis. As more financial markets are being continuously
traded in 24 hour markets, Point and Figure allows all significant price changes to be
recorded without the loss of price change data experienced with other data depiction
techniques. These include such as Open, High, Low, Close charts where intra-period
price movements are lost and even the selection of ‘Open’ and ‘Close’ are arbitrary at
best in continuous markets such as foreign exchange.
3
The relevant literature on Point and Figure is particularly small with only two works
appearing in the academic literature, both being published in German by Hauschild and
Winkelmann (1985) and Stottner (1990),. The remainder of works have been published as
books of varying quality by authors including Aby (1996), Cohen (1960), Dorsey (1995),
Seligman (1962), Wheelan (1954), Zieg and Kaufman (1975) and Davis (1965). These
works are discussed in more detail below.
This paper is designed to bridge that gap between the practitioner and academic literature
by providing a rigourous test of the various Point and Figure chart ‘patterns’ said to
produce profitable trading opportunities. These are tested by mathematically specifying
each of the patterns, then simulating the trades specified by the trading rules on S&P500
futures contracts and reporting the profitability in an EMH framework.
The remainder of this paper is structured as follows. Section 2 discusses the key literature
in the field of Point and Figure and market efficiency. Section 3 provides a computational
specification of Point and Figure as applied in this research. Section 4 defines the trading
signals/rules that were adopted, while Section 5 presents the results and the paper
concludes in Section 6.
II PREVIOUS RESEARCH
The earliest reference in Point and Figure charting appears to be deVilliers (1933), who
claims that the method has “…grown from crude beginnings more than fifty years ago
[and is] …herewith described for the first time” (deVilliers, 1933:7). Assuming this
statement is accurate, it implies that Point and Figure’s usage extends to the mid-late 19th
century. Numerous books have been produced on this topic during the 20th century by
authors including Aby (1996), Cohen (1960), Dorsey (1995), Seligman (1962), Wheelan
(1954), Zieg and Kaufman (1975) and Davis (1965).
Most of these works provide reasonably elementary treatment of the subject and/or
provide largely unstructured methodologies that are unsuitable for rigorous academic
4
journals. Examples of poor methodology include the use of spurious trendlines that have
little a priori value, vaguely defined/subjective chart ‘patterns’ and trade entry/exit
‘rules’ which become so onerous in their specification that they are unlikely to be of
practical value due to the rarity of such complex conditions being met. Just as technical
analysts working with bar charts claimed the existence of patterns that were subjective
and/or poorly specified, such as the only recently quantified Head and Shoulders patterns
(see Osler, 1998), Point and Figure has also attracted its share of essentially subjective
and unreplicable patterns.
Examples are provided in Cohen (1985) who discusses nebulous and ill-defined patterns
including the ‘Inverse Fulcrum’ and the ‘Saucer’ with their vaguely parabolic shapes and
the ‘Compound Fulcrum’ with trading producing two local minima of roughly equal
values. It is suspected that the subjectivity which plagues many popular ‘charting’ works,
including most of those above, have correctly attracted considerable scepticism from
academics requiring standards of replicability and objectivity.
Some research in this area has provided a structured and replicable methodology which
provides a valid testing framework for assessing the profitability of Point and Figure
1
For a definition of Ultra-High-Frequency data see Engle (2000).
2
Note that Australian Interest Rate Futures are quoted as 100 – Yield and so a half point is considerably
smaller in dollar value than that observed in US Interest Rate futures contracts.
5
charting for trading rule researchers. Only two such works examining trading rules using
Point and Figure appear to have been published in refereed finance journals and these
were published in German by Hauschild and Winkelmann (1985) and Stottner (1990).3
Hauschild and Winkelmann (1985) examined several simple Point and Figure trading
rules using daily data on 40 companies listed on German equity markets between 1970
and 1980. Their use of daily data can produce some problems with the calculation of
Point and Figure results. For example, when dealing with Open, High, Low, Close data
inferences/guesses must be made about whether the day’s highest price was traded before
the day’s low to determine whether a price reversal has occurred during that day.
Furthermore, if only closing prices are used then trading activity through the day (which
may have produced a buy/sell signal) is not recorded reducing the accuracy of the
recorded price movements. Therefore these limitations arising from the use of daily data
can achieve only a limited approximation to the more accurate use of intra-day data
which is able to capture all price movements for an asset.4
Hauschild and Winkelmann (1985) did not present results for individual firms and so the
composition of the component results are not available for discussion. On the aggregated
results across all firms the Point and Figure technique was unable to outperform a simple
buy-and-hold strategy for the period.
Stottner (1990) also examined equity markets examining 445 German and overseas
companies. The data set comprised closing data for periods of between 70 months and 14
years prior to the conclusion of the test in February 1989. Stottner (1990) used Point and
Figure charting but in a manner more akin to a simple filter-rule strategy with no
complex pattern assessment. As with Hauschild and Winkelmann (1985), he also found
that Point and Figure produced trading results inferior to a simple buy-and-hold strategy.
3
Both articles gratefully translated by Ralf Becker, an econometrics PhD student at Queensland University
of Technology.
4
The techniques for using daily data with Point and Figure are discussed in most of the books referred to in
this literature review section.
6
The use of the filter rule approach casts some doubt as to the ability to fully assess the
results as an accurate reflection of Point and Figure trading rule performance during the
test period. This is because the technique adopted in Stottner (1990) considers very
simple Point and Figure trading rules without testing the rules that have appeared in much
of the popular Point and Figure literature. As with much filter rule research, the results
presented shared the poor profitability characteristics documented back as far as Fama
and Blume (1966) and Ball (1978).
One of the books published on Point and Figure by Zieg and Kaufman (1975) produced a
methodology capable of being reproduced. This consisted of a well-defined set of eight
buy and eight sell strategies, labelled B1 to B8 and S1 to S8 respectively, complete with
results produced in Davis (1965) discussed at length. Their technique is adopted in the
current study and is fully defined in Section III.
The Davis (1965) study examined daily price data for 1,100 US equities between 1954
and 1964 with remarkable results. Of the eight different buy signals examined, profits
were produced on 71%-92% of trades across the different rules. All eight of the sell
signals examined were profitable in greater than 80% of trades modelled in the
simulation. Claims of such startling profitability demanded a review of his method on
modern markets to see if such consistent profitability is still available to trading
practitioners. This would also require a discussion of the implications for market
efficiency if such results are still able to be replicated in contemporary markets.
As this study deals with trading rules, all profitability should be considered in the context
of the Weak-Form of the Efficient Markets Hypothesis (EMH) proposed by Fama (1970).
As is well known, under this hypothesis consistent profits should not be available where
the only information used is historical prices.
Proponents of technical analysis would argue that historical data does contain
information and therefore all information is not impounded in a security’s price. This
paper relies on trading rules formulated around the Point and Figure methodology with
7
the aim of determining whether the trading rule returns for S&P futures are greater than
zero, so providing economic benefits to traders.
deVilliers (1933) implies that the use of Point and Figure provided substantial benefits
for traders managing data from early ticker-tape machines. This technique is possibly the
first attempt to deal with the very large data sets produced from the use of ultra-high-
frequency (UHF) data, that is where every trade has been recorded for a given financial
instrument. Point and Figure’s ability to reduce UHF data to whatever size price
movement the analyst regards as significant provides computational benefits when
analysing continuously trading markets. These 24 hour markets include foreign exchange
and an increasing number of futures instruments where the imposition of arbitrary
Open/High/Low/Close points may not be suitable.
When the price sequence in Figure 1 is converted into Point and Figure a number of
factors beyond simple Price and Time axes need to be considered in how prices are
recorded. Two variables need to be specified, namely the Points Per Box (PPB) and the
Reversal Size (REV). The PPB determines what level of price sensitivity/significance is
to be recorded in each ‘Box’ and in this example is set at PPB = $1. REV specifies how
many ‘Boxes’ the price needs to reverse by before new price changes are recorded and
REV = 3 in this example.
Some minor methodological variations between authors occur and the specifications here,
while faithful to the original premise, provide a technique suitable for computer-based
processing of source data. Point and Figure relies on the specification of two variables.
Firstly, the number of Points Per Box (PPB) which specifies the coarseness of the data-
filtering such as $0.50, $1.00, $2.00 etc. PPB determines what will be considered a
‘significant’ price change. The second variable is the Reversal (REV) amount. This
determines how many ‘Boxes’ the price must change by to have the movement recorded.
Therefore if PPB = $1.00 and REV = 3 then price must reverse by $3.00 to be recorded.
9
Point and Figure requires data rounding to occur via a series of continuous modulus
operations rounding to the value specified for PPB.5 The input price data are rounded-up
when prices are declining and rounded-down when prices are increasing. The opening
direction of prices, ie falling or rising must first be determined. Some authors will adopt
the first price as the starting point, though this can cause significant computational
difficulties. In this research prices are read from a data file/feed until the remainder of
Price, Pt , divided by PPB equals zero according to equation (2).
Prices continue to be input to establish whether prices are rising or falling on the initial
movement being recorded. Assuming that PPB = $1.00 and REV = 3, then subsequent
prices, Pn , from the data file are input until either equation (3) or (4) is true.
Pn Pt + 3(PPB) (3)
Pn Pt - 3(PPB) (4)
Had equation (3) been satisfied first, all subsequent raw input prices, Pr, read are rounded
down via the modulus operation per equation (5) to produce the filtered price, Pf, rounded
per the Point and Figure methodology.
Pn *
Pf = Pn MOD (Pn , PPB)
(PPB) * *
(5)
PPB −
PPB
* Where MOD (Pn , PPB) = 0
** Where MOD (Pn , PPB) 0
5
Modulus operations involve dividing x by y and reporting the remainder, therefore MOD (11, 3) = 2. Note
that in some programming languages, such as Visual Basic, all decimals must be removed before the
modulus operation is performed or else only the integer part of the expression is evaluated. Therefore, in
this research the evaluation of MOD (11.65, 3) would require both x and y to be multiplied by 10n until the
decimals (thus 10n = 102 ) are removed so producing the expression MOD (1165, 300) = 3.88 to be
10
Had equation (4) been satisfied initially instead of equation (3), the data would have
needed to be rounded-up per equation (6).
Pn *
Pf = Pn MOD (Pn , PPB)
(PPB) * *
(6)
1 + PPB −
PPB
* Where MOD (Pn , PPB) = 0
** Where MOD (Pn , PPB) 0
Data continues to be read according to the modulus operations above until Pn triggers the
next entry. If prices were increasing, and so equation (3) was initially satisfied, the next
price recorded is where Pn (P f + PPB) or price declines from the highest point in that
movement where Pn [Pf – REV(PPB) ].
Had prices been declining, and so equation (4) was satisfied, prices would be continued
to be rounded-up per equation (6). Prices would then continue to be read until either Pn
(Pf – PPB) or price increases from the lowest point in that movement so that Pn [P f +
REV(PPB)]. These processes are repeated until all data are exhausted.
In summary, the initial price and direction must be established as the subsequent
calculations may display sensitivity to the initial starting price and starting on another
date may lead to differing results arising from differences in the initial direction as shown
in equations (3) and (4). Once the initial price is established, the data is read into two
distinct loops for rounding-down (when prices are falling) and rounding-up (when prices
are rising). This raises a methodological concern that two analysts beginning with
different starting points would end up with differences in the way the filtering is
conducted. It is suggested that any researchers dealing with this methodology should
clearly state their technique adopted for subsequently empirical work.
evaluated to produce the true remainder. A complete numerical example of the procedures adopted here are
shown in Appendix III.
11
Little consensus arises between the authors as to what PPB should be set at and this is left
to the analyst to determine in the context of the price volatility of the asset being
examined and the investment horizon. This is not so with REV where most of the books
described above applies REV = 3, eg Dorsey (1995) states that “…we never deviate from
the 3-point reversal method … [though] you may want to choose other reversal points”.
Therefore it may be assumed that other values, eg PPB = 5, might have usage as little
evidence has been presented to date. Consequently this study will include a sensitivity
analysis across a range of different PPB and REV values for the futures contracts
considered.
One concern about the Point and Figure technique arises from what price is recorded
when rounding input data. Assume that a new entry should be made on a Point and
Figure chart and a buy signal (explained in the next section) is generated when a price of
105.0 is reached. If the price trades at 105.0 then the entry could be validly made and the
trade simulated at that price. But if the market is more volatile and the price generating
the entry at 105.0 actually traded at 105.25, then the long position assumed to be taken at
105.0 would overstate profits.
Analogous to this problem is when ‘gapping’ in the price series occurs. That is where for
example the market closes at 104.5 and re-opens the next day at 106.0, but a buy signal
was to be generated at 105.0. This also overstates profits because the trade is simulated to
occur at 105.0 but could not have been taken until 106.0.
In this research, efforts have been made to address this problem by having the price that
caused the movement to be recorded, rather than just the rounded Point and Figure value.
Samples of converted Point and Figure data output have shown that this issue has been
adequately dealt with by the use of ‘actual’ price rather than simply ‘rounded’ price, but
some gaps may have escaped the detection process. This does not appear to have
significantly disturbed the results here for most trading rules, but is an issue that must be
given serious consideration by other researchers proposing studies in this area.
12
This section outlines the trading rules adopted in assessing the usefulness of Point and
Figure charting as a market timing tool. Point and Figure charts are constructed in such a
way as to readily permit the analysis of the forecasting ability of chart patterns.
The poor replicability of pattern recognition has lead to little treatment in the literature,
although some recent replicable methodologies have been provided by Osler (1998) and
Lo (2000). With some Point and Figure signals however, their mathematical specification
can be simplified into simple logical Boolean statements. The trading rules adopted here
were applied in Davis (1965) and are reproduced below labelled as buy signals B1 to B8
and as sell signals S1 to S8 with an intuitive explanation as to their rationale. The
Boolean specification is provided in Appendix B.
The Double Top (Double Bottom) formation is, by definition, the most widely observed
trading pattern in Point and Figure as all of the more sophisticated patterns discussed
below must contain this basic pattern. The formation occurs by prices rising above
(below) the previously established highest price. It implies that prices trading above
(below) a previous high (low) suggest that the market is subject to an increase in demand
(supply) beyond the local maxima (minima) and that the stronger demand (supply) will
persist. Consequently the continued buying (selling) should cause prices to increase
(decrease) so producing a profitable trading opportunity.
The Double Top with Rising Bottom (Double Bottom with Declining Top) formation
extends the condition in B1 (S1) by adding the requirement that the previous low (high)
is higher (lower) than its preceding low (high) as measured on the columns of O’s (X’s).
The rationale of this formation may be that the presence of higher highs (lower lows) and
higher lows (lower highs) indicates more pronounced and sustained demand (supply) has
emerged in the market and that prices will continue to reflect this increasing demand
(supply). This would suggest the expected persistence of rising (falling) prices so
producing a profitable long (short) position.
The Breakout of Triple Top (Breakout of Triple Bottom) formation suggests that prices
have traded to a previous high (low) on two separate occasions, only to be met with
supply (demand) at that price level. On the third occasion, demand (supply) has been has
been strong enough to satisfy sellers (buyers) at that level and the increased demand has
been sufficient to cause a price increase (decrease). The implication is therefore that the
demand (supply) will continue to be present and that prices will continue to rise (fall)
producing a profitable long (short) trading opportunity.
14
The Ascending Triple Top (Descending Triple Bottom) extends on the Breakout of Triple
Top (Breakout of Triple Bottom) by requiring the lows (highs) shown in the columns of
O’s (X’s) to be higher (lower) and also the highs (lows ) indicated by the columns of X’s
(O’s) to all be rising (falling). Again, the inference here is that the sustained demand
(supply) indicated by the persistently rising (falling) prices will continue to produce a
profitable long (short) trading opportunity.
As with the Breakout Triple Top (Breakout Triple Bottom), the Spread Triple Top
(Spread Triple Bottom) the formation infers that supply (demand) has previously entered
that market at a given price. The rising (falling) of prices beyond the previously
determined high (low) suggests that the supply (demand) has now been satisfied and
sufficient demand (supply) has now emerged to cause prices to continue to increase
(decline) to new levels. Consequently, the expected increase in demand (supply) should
produce profitable long (short) trading opportunities.
15
The ‘triangle’ formation and its many variants such as bullish/bearish triangles,
rising/falling wedges etc have long appeared in technical analysis (see Edwards and
Magee, 1961). The formation implies that a lack of information into the market has led to
neither supply or demand dominance and consequentially no direction in prices.
The triangle pattern may imply that inventory readjustment is occurring rather than price
being information-driven. The ‘Breakout’ of the triangle would then suggest that either
new information has arrived in the market or a significant inventory readjustment is
occurring. The analyst may then infer that prices will continue to move in the same
direction as the price breakout from the triangle's apex and be positioned long (short).
The rationale for why a buy (sell) level appears where it does is not intuitively clear. The
use of support/resistance lines have appeared widely in practitioner literature since
Edwards and Magee (1961) and are applied here in Point and Figure. The basic idea may
16
As with signals B7:S7, the use of a trendline is employed for the trading rule. In the case
of B8, a breach of the trendline suggests that a reversal in price direction has occurred
and the trader should open a long position to capitalise. S8 suggests that, while prices are
still moving lower, a more vigorous supply situation has emerged and the trader should
hold a short position to capitalise on the expected continuance of price decline.
When considering the above trading rules/patterns, all cases except B5, S5, B7 and S8
decompose into the simple Double Top/Bottom formation. It is expected that the
additional signal conditions, eg Triple Top, grew to minimise the transaction costs
associated with frequent trading where almost every change in price direction recorded
would generate a trading signal from B1:S1.
17
For this study all orders are assumed to be ‘Stop’ orders where a buy/sell signal is
produced. This means that trades are made during one of the up/down movements
without waiting for a closing price as relied on in much of the technical trading rule
literature. All open positions are closed at the last data point for each year.
In accordance with numerous trading rule studies transaction costs are modelled into the
results. These have been set at $100 round-turn per futures contract traded in accordance
with comparable amounts in Lukac et al. (1988), Anderson (1997), Babcock (1989,
Bilson and Hsieh (1987), Boothe and Longworth (1986), Lukac and Brorsen (1989),
Sweeney (1986) and Taylor (1993).
Zieg and Kaufman (1975) suggest that positions should be taken corresponding to each
trading signal generated by the trading rules. Consequently, the number of contracts taken
on each trading signal is one contract, although much larger positions may be
accumulated from successive buy or sell signals. For each individual signal pair, eg
B4:S4, the long (short) position generated by the signal is closed out by the first
occurrence of S1:B1 respectively in accordance with Davis (1965). This exit strategy is
used because some of the rarer signals, such as B8:S8, may not get the opposing signal
for that pair (ie S8:B8 respectively) and need some other position exit requirement.
Given the similarity between signals B1:S1 and the other trading rules, more than one
position may be initiated at the same price due to the overlap of signals. Similarly, if
signal B1 is acted upon, and another B1 signal is generated, then two positions will be
held. Subsequent positions will also be taken and no limits on the position size have been
imposed for this simulation. The trading rules do not require the specification of a trade
6
These values were selected as part of a broader doctoral thesis examining Point and Figure across
numerous futures markets and using common dollar values across different futures contracts.
18
exit signal as an opposing trade entry signal will cause termination of all long (short)
positions and new short (long) positions to be taken. All positions are closed on the last
price of the final day for each year.
Results are presented here for the S&P 500 futures contract between 1990 and 1998. The
S&P500 futures contract value is calculated as 250 times the index, giving a dollar value
of $250 per ‘big point’. The spot contract, or nearest contract to expiry, has been used to
avoid liquidity problems that may be present in distant contracts. The futures contract
price series were adjusted to remove any artificial profits/losses on contract expiration7 .
This section is organised into sections discussing the trading rule profitability. It
discusses the average annual performance and the performance across trading rules. The
performance in contemporary index futures markets is then critically assessed against the
earlier documented performance in Zieg and Kaufman (1975). It concludes by discussing
the implications for market efficiency.
7
Contract Rollover is performed automatically via the ‘Autoroll’ technique in the data extraction software
from Tick Data Inc. The spot contract is automatically ‘rolled’ into the next contract when volume in the
following contract exceeds the volume in the expiring contract. The price differential on rollover date is
removed by adjusting all subsequent prices by the differential amount to reflect how a trader would, for
example, roll an long position by selling the position in the expiring contract and simultaneously re-
opening the long position in the subsequent contract expiry. Ma et al. (1992), found that the S&P500
futures were robust across rollover methods and the rollover method used here should not produce
significant impacts on the data examined.
19
Table 1 provides a summary of the trading result annual averages by trading rule. A more
detailed set of results are provided in Appendix B outlining year-by-year performances of
each trading rule.
Table 1: Trading Rule Result Summary – Total Profitability
3BR 4BR 5BR 3BR 4BR 5BR
Strategy
$100PPB $100PPB $100PPB $200PPB $200PPB $200PPB
B1:S1
NumTrades 18,278 9,128 10,110 4,012 2,393 1,685
%Profitable† 43 43 40 41 43 43
Gross Profit 1,463,200 1,393,772 1,248,537 588,800 825,888 630,724
Net Profit -364,600 480,972 237,537 187,600 586,588 462,224
B2:S2
NumTrades 8,433 5,509 6,168 2,286 1,494 1,067
%Profitable† 42 41 38 38 41 40
Gross Profit 708,800 722,936 629,550 434,000 410,974 353,849
Net Profit -134,500 172,036 12,750 205,400 261,574 247,149
B3:S3
NumTrades 2,201 891 908 504 144 79
%Profitable† 48 42 43 45 54 40
Gross Profit 331,800 177,873 192,062 83,200 101,349 35,775
Net Profit 111,700 88,773 101,262 32,800 86,949 27,875
B4:S4
NumTrades 5,517 3,540 3,933 1,498 999 705
%Profitable† 41 41 37 39 40 42
Gross Profit 381,800 454,561 328,651 180,200 262,537 265,011
Net Profit -169,900 100,561 -64,649 30,400 162,637 194,511
B5:S5
NumTrades 429 143 170 90 34 26
%Profitable† 43 48 47 46 56 54
Gross Profit 18,400 37,148 43,925 14,800 23,000 16,800
Net Profit -24,500 22,848 26,925 5,800 19,600 14,200
B6:S6
NumTrades 6 6 0 0 2 0
%Profitable† 33 33 0 0 50 0
Gross Profit 100 -300 0 0 400 0
Net Profit -500 -900 0 0 200 0
B7:S7
NumTrades 49 24 22 19 5 2
%Profitable† 52 54 27 58 40 0
Gross Profit 11,000 9,375 -3,025 5,200 8,350 -400
Net Profit 6,100 6,975 -5,225 3,300 7,850 -600
B8:S8
NumTrades 63 27 16 21 9 3
%Profitable† 46 45 50 48 66 67
Gross Profit -500 7,200 2,625 9,600 5,525 6,000
Net Profit -6,800 4,500 1,025 7,500 4,625 5,700
Totals
NumTrades 34,977 19,302 21,355 8,477 5,124 3,566
%Profitable 43 42 39 40 42 42
Gross Profit 2,914,600 2,802,565* 2,442,325** 1,315,800 1,638,023** 1,307,759*
Net Profit -583,100 872,365 306,825** 468,100 1,125,623** 951,159*
†
‘% Profitable’ results for all trades in that category
*t-test of average annual profit significant at 0.10 level **t-test of average annual profit significant at 0.05 level
B1 Double Top S1 Double Bottom
B2 Double Top With Rising Bottom S2 Double Bottom With Declining Top
B3 Breakout of Triple Top S3 Breakout of Triple Bottom
B4 Ascending Triple Top S4 Descending Triple Bottom
B5 Spread Triple Top S5 Spread Triple Bottom
B6 Upside Breakout Of Bullish Triangle S6 Downside Breakout of Bearish Triangle
B7 Upside Breakout Above Bullish Resistance Line S7 Downside Breakout Below Bullish Support Line
B8 Upside Breakout Above Bearish Resistance Line S8 Downside Breakout Below Bearish Support Line
Table 1 provides a summary of trading performance outlining the Number of Trades
(NumTrades) each trading rule undertook, the percentage of these trades that were
profitable before transaction costs (%Profitable), the Gross Profit (dollar profit/loss
before any allowance for transaction costs) and Net Profit adjusted for transaction costs is
calculated as [Gross Profit – (NumTrades x $100)].
As expected, signals B1:S1 produced the largest number of trades and the number of
trades for subsequent trading rules declined as entry/exit conditions became more
restrictive. One important methodological difference adopted in this paper, ie using the
price that triggered the Point and Figure entry to be recorded rather than simply the
rounded value, has meant that signal B6:S6 recorded very few trades. This result would
arguably be different had simply rounded values been used, but would have led to
significant profit reporting inaccuracies.
All PPB and REV levels tested during the trading period produced positive net profits
except for the smallest filtering level tested, namely PPB = $100 and REV = 3. Although
these values produced positive Gross Profits of $2,914,600 the large number of
transactions (34,977) negated the economic value of such a strategy.
The highest net profit recorded during the test period was produced with PPB = $200 and
REV = 4. Most trading rule variable selection represents some form of trade-off between
a large number of transactions with low average profit per trade (often failing to cover
transaction costs) and a lower number of transactions with higher average profit per trade
(often requiring greater funding costs as larger unrealised losses may need to be funded).
This balance between gross profitability ($1,638,023) and number of annual transactions
(5,124) produced superior net profitability ($1,125,623) during the test period for PPB =
$200 and REV = 4.
In accordance with Brock et al. (1992), a t-test was performed examining the differences
in the mean returns of the trading rules and the zero-expected return for the S&P futures
22
contract.8 The net profits generated in the simulation were significantly different from
zero for the following filtering levels (REV) and Points Per Box (PPB): PPB = $100 for
the REV values of 4 and 5, while PPB = $200 also produced significant profits for REV
values of 4 and 5.
The popular value of REV = 3 failed to produce statistically significant profits for either
PPB = $100 or PPB = $200. The reasons for this are not immediately apparent but may
alter with the use of higher PPB values. While excessive trading and the resulting higher
transaction costs may explain these results with some trading rules, the Gross Profits (ie
before any allowance for transaction costs) failed to produce statistically significant
results and does not produce any intuitively appealing reason for this result.
All PPB and REV values produced Gross Profits, but all with percentage of profitable
trades less than 50%, and with none producing profits on greater than 42% of trades. It
can be concluded that the average profit on successful trades were higher than the
average loss on the losing trades. This suggests that the trading rules considered were
able to mechanistically apply the old traders adage of ‘letting profits run and cutting
losses short’.
Table 1 examined the average of rules B1…B8/S1…S8 across all years but, as with many
averages, they may fail to provide sufficient information for meaningful conclusions to
be drawn. Table 2 outlines the performance of all trading rules by year providing a
8
The t-statistic for the annual profitability was calculated as,
µr − µ
(σ 2 / N + σ 2 / N r )
where ìr and Nr are the mean dollar return and the number of years in the test respectively and ì and N are
the zero expected return and number or years in the test.
23
significantly different picture of the performance across different PPB and REV levels
reported in Table 1.
Table 2: Trading Rule Result Summary – Annual Totals For All Trading Rules
Strategy 1990 1991 1992 1993 1994 1995 1996 1997 1998 Totals
PPB = $100
REV = 3
NumTrades 1,213 976 719 635 845 969 2,765 11,719 15,136 34,977
%Profitable 39 42 39 41 40 44 41 43 44 43
Gross Profit 38,300 93,700 -8,600 -41,500 3,800 94,700 234,700 1,036,500 1,463,000 2,914,600
Net Profit -83,000 -3,900 -80,500 -105,000 -80,700 -2,200 -41,800 -135,400 -50,600 -583,100
PPB = $100
REV = 4
NumTrades 717 573 423 377 514 568 1,610 5,231 9,289 19,302
%Profitable 41 37 40 40 40 41 42 42 42 42
Gross Profit 87,111 114,871 1,162 -19,550 57,112 59,136 281,223 985,300 1,236,200 2,802,565
Net Profit 15,411 57,571 -41,138 -57,250 5,712 2,336 120,223 462,200 307,300 872,365
PPB = $100
REV = 5
NumTrades 424 284 256 348 393 1,131 3,718 6,329 8,472 21,355
%Profitable 39 29 41 38 38 41 43 39 38 39
Gross Profit 81,175 -20,387 15,613 63,300 12,012 223,537 679,675 702,600 684,800 2,442,325
Net Profit 38,775 -48,787 -9,987 28,500 -27,288 110,437 307,875 69,700 -162,400 306,825
PPB = $200
REV = 3
NumTrades 319 261 164 152 261 247 769 2,284 4,020 8,477
%Profitable 33 36 33 43 43 36 43 43 40 40
Gross Profit -61,600 22,400 -52,800 3,400 31,600 -19,400 175,800 385,800 830,600 1,315,800
Net Profit -93,500 -3,700 -69,200 -11,800 5,500 -44,100 98,900 157,400 428,600 468,100
PPB = $200
REV = 4
NumTrades 184 147 94 92 112 158 445 1,360 2,532 5,124
%Profitable 41 37 33 44 51 41 47 44 41 42
Gross Profit 35,412 77,150 -2,537 -4,813 84,425 41,637 346,949 458,000 601,800 1,638,023
Net Profit 17,012 62,450 -11,937 -14,013 73,225 25,837 302,449 322,000 348,600 1,125,623
PPB = $200
REV = 5
NumTrades 116 100 69 52 85 105 290 967 1,782 3,566
%Profitable 42 28 50 27 46 34 45 43 42 42
Gross Profit 19,674 26,800 43,749 -32,037 38,237 19,962 259,824 391,350 540,200 1,307,759
Net Profit 8,074 16,800 36,849 -37,237 29,737 9,462 230,824 294,650 362,000 951,159
Totals
NumTrades 2,973 2,341 1,725 1,656 2,210 3,178 9,597 27,890 41,231 92,801
%Profitable 39 38 39 40 41 41 42 42 41 41
Gross Profit 200,072 314,534 -3,413 -31,200 227,186**419,572* 1,978,171**3,959,550** 5,356,600** 12,421,072
Net Profit -97,228 80,434 -175,913 -196,800 6,186 101,772 1,018,471** 1,170,550* 1,233,500* 3,140,972
†
‘% Profitable’ results for all trades in that category
*t-test of PPB, REV profit significant at 0.10 level **t-test of PPB, REV profit significant at 0.05 level
B1 Double Top S1 Double Bottom
B2 Double Top With Rising Bottom S2 Double Bottom With Declining Top
B3 Breakout of Triple Top S3 Breakout of Triple Bottom
B4 Ascending Triple Top S4 Descending Triple Bottom
B5 Spread Triple Top S5 Spread Triple Bottom
B6 Upside Breakout Of Bullish Triangle S6 Downside Breakout of Bearish Triangle
B7 Upside Breakout Above Bullish Resistance Line S7 Downside Breakout Below Bullish Support Line
B8 Upside Breakout Above Bearish Resistance Line S8 Downside Breakout Below Bearish Support Line
Table 2 reveals that the use of Point and Figure trading rules produced very mixed profits
when minor trends and lower volatility were observed in the early-mid 1990s though
higher profits in the latter part of the 1990s. Table 3 reveals the daily return and standard
deviation characteristics observed during the test period.
Once the higher volatility was observed in the late 1990s substantial profits were
produced by the trading rules applied with Gross Profits in excess of $1,000,000 being
reported. Had it not been for the last three years of the sample data, the profitability of
Point and Figure in contemporary stock index futures market would have been
questionable for practitioners and within academic expectations of market efficiency.
To explain the potential for volatility and returns to explain the drastic shifts in
profitability across time, various simple regressions were conducted on the data in Table
3 to determine the impact of volatility and returns on Point and Figure profitability.
Returns, r, were measured by two methods. These were the daily percentage returns, ie r%
= (Pt – Pt-1 )/P t and the daily change in point value, ie r = Pt – Pt-1 . The volatility,
measured by the standard deviation, was also measured for r% and r. These values were
then considered across time for various performance data such as Gross and Net Profits
and the number of trades signalled in the simulation.
26
The results of the regression analysis produced some highly significant values of
Pearson’s R2 during the sample period where the independent and dependent variables
are shown respectively for each row. The results are shown in Table 4 below.
Table 4 clearly shows some of the very strong determinants of profitability of Point and
Figure trading rules during the test period. The higher the standard deviation of simple
point value returns, the higher the number of trades undertaken by the trading rules
generating an R2 of 0.99. This translated in higher gross profits producing an R2 of 0.98
where profitability was not diminished by transaction costs. From these results it may be
assumed that the increase in volatility triggered more trades, and that these trades
generally produced positive gross profits.
So it appears that the greatest profitability has arisen through the higher volatility on the
simple number of points, rather than percentage volatility. This is hardly surprising given
that speculators will not profit from percentage gains per se, but from the total dollar
value of the change in the S&P500 Index futures contract.
This of course raises questions about the future profitability of Point and Figure trading
rules as the S&P rises to higher levels in the future. The tables above would, at first
glance, suggest that profitability should increase as the S&P rises. The increase in
profitability may well be offset by higher transaction costs as the execution of market
orders occur at larger bid/ask spreads accompanying the rise in S&P value.
In accordance with numerous trading rule studies, the results are compared with a Buy-
And-Hold control.9 The Net Profit results are shown in Table 5 compared with the Buy-
And-Hold control.
As the trading rules adopted allow more than one futures contract to be acquired it
becomes difficult to really compare the holding of a single futures contract against a
28
multiple position trading strategy. Consequently only limited information can be drawn
from Table 5 which has been included to show that some form of positive drift was
present in the futures contract. The contribution of the drift shown by a simple Buy-And-
Hold strategy did not correlate as strongly as other measures with the profitability of
Point and Figure rules.
In the context of market efficiency, the results would present a mixed view of the
performance of Point and Figure during the test period. Peterson and Leuthold (1982)
suggested that a z-test should be conducted on the reported net profit results against the
zero-expected excess return benchmark of EMH. Using a t- test, the results rejected EMH
at the 0.05 level for PPB =$200, REV = 4 and PPB =$200, REV = 5 during the test
period. In contrast though, it should be noted that the elusive ‘consistent excess returns’
have remained just that as even those parameters rejecting EMH did produce losses in
some years.
It should also be noted that the observed profitability in the latter years of the test does
raise some doubt as to how well the EMH was able to describe the S&P futures contract
during the test period. The presumed popularity of Point and Figure charting given its
prominence in professional market analysis software would suggest that some market
participants may have derived profits from futures trading despite EMH’s theoretical
predictions of zero-expected returns.
With respect to the reported profitability in Davis (1965) evidence was presented
showing the above trading rules had the percentage of profitable trades generally over
80% for all trading rules/chart patterns – a primary motivating factor for this research.
Such high levels of profitable trade percentages were not reflected in this study where the
percentage of profitable trades were almost exclusively less than 50%.
9
Although it should be noted that Peterson and Leuthold (1982), regarded the Buy-And-Hold as useful as a
Sell-And-Hold strategy under the assumption that futures markets price series largely reflect a drift-less
random walk.
29
No clear answer emerges as to the reasons for such differences. This paper merely sought
to test the trading rules in a contemporary dataset in an efficient market framework.
VI CONCLUSION
This study has tested the plausibility of trading rules using specific buy/sell signals
accompanying Point and Figure charting, claimed to be one of the oldest practitioner
techniques with origins in the 19th century. The trading rules, in contrast with other
popular ‘chart patterns’, provided a replicable trading rule methodology and were applied
to the S&P 500 futures contracts traded between 1990 and 1998.
The results found that profits were available to speculators on the S&P 500 futures
contract in contrast to the zero-expected excess returns available to speculators under the
efficient market hypothesis (EMH). This study, as with trading rule studies generally,
was unable to reject EMH in its strictest sense as profits were not consistently available
to the speculator. This is because loss generating years were reported - profits were only
available on aggregate.
Most of the profits appear to have derived from the relatively high volatility levels, on a
per-point basis rather than a percentage basis, producing an R2 close to one between S&P
30
500 futures point volatility and trading rule profits. During years when volatility was low,
profit results were mixed, although the higher daily volatility observed in the late 1990s
meant that large profits were available to speculators using Point and Figure trading rules.
31
This appendix provides a numerical example for the modification of data using the
structure of Point and Figure charting. Assume that we set the values of the Points Per
Box (PPB) = $1.00 per box and the Reversal Size (REV) = 3. Given the following
hypothetical price data in Panel A-A (reading left to right) set the data is transformed by
the following steps.
Panel A-A: Hypothetical Data Set for Point and Figure Chart
100 101 102 101 99 98 97 98 99 100
99 100 101 102 101 100 101 99 100 99
Step 1: Ensure that the first data point (100) can be divided equally, ie with no remainder,
by the value set for PPB. As $100/$1 produces no remainder, $100 is used as the starting
value. Had the first value equalled $100.25, data would have continued to be read until a
remainder of zero was produced.
Step 2: At this point it is unknown whether the data is rising or falling and so no entry is
made on the Point and Figure chart until a value of the first data point plus/minus the
REV level is established, that is Price = $100 ± $3. Data is read until the price $97 is
input. At this point the first entry on the Point and Figure chart can be entered as shown
in Figure A-1.
Step 3: Having established the initial direction, data continues to be read until either a
price of $96 is read, causing an ‘O’ to be recorded at $96, or the price of the local minima
32
plus REV times PPB is read, that is $97 + (3 x $1). As a price of $100 is encountered
before $96, the Point and Figure chart now appears as in Figure A-2.
As prices are now rising and the last entry recorded was $100, data continues to be read
until either a price of $101 is read, causing an ‘X’ to be recorded at $101, or the price of
the local maxima minus REV times PPB is read, that is $100 - (3 x $1). As a price of
$101 is encountered before $97, the Point and Figure chart now appears as in Figure A-3.
Similarly, a price of either $102 or $98 will trigger the next entry. As the following price
read is $102, the Point and Figure chart now appears as sown in Figure A-4.
Now that the local maxima is $102, data continues to be read until either $103 or $99 is
read. As the next significant price, according to the values used for PPB and REV, to be
read is $99, the final Point and Figure chart now appears as shown in Figure A-5.
As all data is now exhausted, the final Point and Figure chart is shown in Figure A-5.
Data size is dramatically reduced from 20 original data points down to 3 points. Point and
Figure charting totally dispenses with time on the x-axis and time data is totally lost in the
filtration process. Therefore it does not record how long it took for an entry to be made,
but merely the movements in price.
The mathematical description of Point and Figure used in this paper draws extensively on
the Modulus operation, that is dividing x by y and reporting the remainder. As securities
rarely trade exclusively at even dollar amounts, the modulus operation is used to detect
when the price read is evenly divisible by PPB so that the software detects when entries
are to be made or not. When using PPB = $1, this effect is not as apparent as a value of
PPB = $3. In this example The first recorded price would have been $99, ie MOD (Price,
PPB) being MOD ($99, $3) producing a remainder of zero, and no further entries would
have occurred.
34
This section outlines the specification of the trading rules per the structure of Boolean
Logic where conditions evaluate as either True or False. They generally take the form of
an ‘IF…THEN…ELSE’ statement where IF x evaluates as true, THEN conduct operation
y. If condition x evaluates as False, the ‘Else’ statement requires that operation z should
be conducted.
The conditional statements also require the data filtered according to the Point and Figure
methodology be thought of as existing in an 2-dimensional array. This is demonstrated in
Figure B-1 as follows.
Figure B-1 shows how the data can be interpreted with some consistency when applying
the trading rules accompanying Point and Figure. The data in the current move is held in
array element [0], while the previous movement is recorded in array element [-1] etc. The
size of each movement is considered to be the highest price in array element n minus the
lowest price in array element n. Therefore, array element [-2] has a value of 3 as does
element [-1] etc.
All conditional statements described below take a Boolean form separated by curly
brackets for each part of the condition. The form adopted in this rule description is then;
IF {abc = True} THEN {undertake action xyz}
35
If condition ‘abc’ returns false, the next datum is read until the condition returns true and
either a trade is initiated or liquidated.
All orders are treated as ‘stop-loss’ orders, meaning that they are executed at market. In
this simulation the price at which the trade occurs is the actual price at which the Point
and Figure entry was generated. Consequently, some trades will occur at twentieth’s of
one point if that was the value that caused the Point and Figure entry to be recorded. See
Footnote 3 for more explanation of this point.
X ⇐ BUY X
X X O X O
X O X O X O
X O X O O
O O ⇐ SELL
X ⇐ BUY X
X X X O X
O X O X X O X O
O X O X O X O
O X O O O
O O ⇐ SELL
X ⇐ BUY X X
X X X O X O X O
X O X O X O X O X O
X O X O X O O O
O O O ⇐ SELL
X ⇐ BUY X
X O X O X O X
X X O X O X O X O
X O X O O O X O
X O X O O
O O ⇐ SELL
Buy Signal – B4 IF {(low[-1] > low[-3]) AND (high[-2] > high[-4]) AND
(Price = High[-2] + (PPB))}
THEN
{Buy High[-2] + (PPB) Stop;}
Sell Signal – S4 IF {(high[-1] < high[-3]) AND (low[-2] < low[-4]) AND
(Price = Low[-2] - (PPB))}
THEN
{Sell Low[-2] - (PPB) Stop;}
X ⇐ BUY X X X
X X X X O X O X O
X O X O X X O X O X O X O
X O X O X O X O X O X O O
O X O X O X O O O
O O O O ⇐ SELL
37
Bibliography
Aby, C. D. J. (1996) Point & Figure Charting: The Complete Guide, Traders Press Inc,
Greenville, SC, USA.
Anderson, J. A. (1997), Trading Rules, Parameter Selection & Portfolio Effects In
Australian Ten Year Bond Futures, Master of Business, School of Business,
Southern Cross University, Lismore.
Anderson, J. A. (1999), In 1999 Asia-Pacific Finance Association ConferenceMelbourne
Australia.
Babcock, B. J. (1989), The Dow-Jones Irwin Guide To Trading Systems, Richard D.
Irwin Inc, Illinois USA.
Ball, R. (1978), Filter Rules: Interpretation of Market Efficiency, Experimental Problems
and Australian Evidence, Accounting Education, November, 1-17.
Bilson, J. F. O. and Hsieh, D. A. (1987), The Profitability of Currency Speculation,
International Journal of Forecasting, 3, 115-130.
Boothe, P. and Longworth, D. (1986), Foreign Exchange Market Efficiency Tests:
Implications of Recent Empirical Findings, Journal of International Money and
Finance, 5, 135-132.
Brock, W., Lakonishok, J. and LeBaron, B. (1992), Simple Technical Trading Rules and
the Stochastic Properties of Stock Returns, Journal of Finance, XLVII, 1731-1764.
Cohen, A. (1960), The Chartcraft Method of Point and Figure Trading, Chartcraft,
Larchmont, NY, USA.
Cohen, A. (1985), In The McGraw-Hill Handbook of Commodities(Eds, Sabin, W.,
Ulwick, C. a. and Kornbluth, G.) McGraw-Hill, USA.
Davis, R. E. (1965), Profit and Probability, Dunn & Hargitt's Financial Service, West
Lafayette, Indiana, USA.
deVilliers, V. (1933), The Point & Figure Method of Anticipating Stock Price
Movements, Windsor Books - Reprint of 1933 Edition, NY.
Dorsey, T. J. (1995), Point and Figure Charting: Essential Applications for Forecasting
and Tracking Market Prices, John Wiley & Sons Inc, USA.
Edwards, R. D. and Magee, J. (1961), Technical Analysis of Stock Market Trends, John
Magee, Springfield, Mass. USA.
Engle, R. F. (2000), The Econometrics of Ultra-High-Frequency Data, Econometrica, 68,
1-22.
Fama, E. F. (1970), Efficient Capital Markets: A Review of Theory & Empirical Work,
Journal of Finance, , 383-417.
Fama, E. F. and Blume, M. (1966), Filter Rules and Stock Market Trading Profits,
Journal of Business, 39 (Special Supplement).
Hauschild, K. and Winkelmann, M. (1985), Kapitalmarketeffizienz und Point and Figure
Analyse, Kredit Und Kapital, 18.
Lo, A. (2000), Foundations of Technical Analysis: Computational Algorithms, Statistical
Inference, and Empirical Implementation, Journal of Finance, 55.
Lukac, L. P. and Brorsen, B. W. (1989), The Usefulness of Historical Data in Selecting
Parameters for Technical Trading Systems, The Journal of Futures Markets, 9, 55-
65.
46